This repository helps you generate a veritable cornucopia of persian licence plates. You're almost done with this part, training your own CNN would be the next step!
Generated sample:
For the sake of ease, generated licence plates come with their annotations:
Generated sample with perspective transformations:
On top of all that, it automatically generates associated xml files in pascal voc format:
Sample neural network trained on this dataset:
Model zoo faster_rcnn_inception_v2_coco
Because of oop pattern you won't need to make significant changes unless you want something structurally different!
*Note: if you have trouble installing jsonnet on windows
use win_conf.json instead and remove jsonnet dependencies
(For installing jsonnet on windows run: "pip install jsonnet-binary")-
Got all requirements installed:
pip3 install --upgrade -r requirements.txt -
Tweak configuration file: (optional)
- project_configurations.jsonnet
-
Out-of-the-box data generation:
- With default parameters:
$ python3 main.py - Override config parameters:
$ pyhton3 main.py --num_out_img 10000
$ pyhton3 main.py --apply_misc_noise False
$ pyhton3 main.py --apply_dirt False
$ pyhton3 main.py --output_directory 'output'(doesn't already exist)
$ pyhton3 main.py --img_per_package 8000
- With default parameters: